MCP Memory

import { v4 as uuidv4 } from "uuid"; // Minimum similarity score for vector search results const MINIMUM_SIMILARITY_SCORE = 0.5; /** * Generates vector embeddings from text using Cloudflare's AI model * * @param text - The text to convert into vector embeddings * @param env - Environment containing AI service access * @returns Promise resolving to an array of numerical values representing the text embedding */ async function generateEmbeddings(text: string, env: Env): Promise<number[]> { const embeddings = (await env.AI.run("@cf/baai/bge-m3", { text, })) as AiTextEmbeddingsOutput; const values = embeddings.data[0]; if (!values) throw new Error("Failed to generate vector embedding"); return values; } /** * Stores a memory in Vectorize with its vector embedding and returns the generated ID * @param text - The memory content to store * @param userId - User ID to associate with the memory (used as namespace) * @param env - Environment containing Vectorize and AI services * @returns Promise resolving to the unique memory ID */ export async function storeMemory( text: string, userId: string, env: Env, metadata: Record<string, string> = {}, memoryId: string = uuidv4() ): Promise<string> { // Generate embedding const values = await generateEmbeddings(text, env); // Store in Vectorize await env.VECTORIZE.upsert([ { id: memoryId, values, namespace: userId, metadata: { content: text, ...metadata }, }, ]); return memoryId; } /** * Search for memories by semantic similarity * @param query - The query to search for * @param userId - User ID to search within (used as namespace) * @param env - Environment containing Vectorize service * @returns Promise resolving to an array of memories matching the query */ export async function searchMemories( query: string, userId: string, env: Env ): Promise<Array<{ content: string; score: number; id: string; metadata?: Record<string, string> }>> { // Generate embedding for query const queryVector = await generateEmbeddings(query, env); // Search Vectorize const results = await env.VECTORIZE.query(queryVector, { namespace: userId, topK: 10, returnMetadata: "all", }); if (!results.matches || results.matches.length === 0) { return []; } // Process results const memories = results.matches .filter((match) => match.score > MINIMUM_SIMILARITY_SCORE) .map((match) => { // Ensure content is a string let content = "Missing memory content"; if (match.metadata && typeof match.metadata.content === "string") { content = match.metadata.content; } else if (match.id) { content = `Missing memory content (ID: ${match.id})`; } return { content, score: match.score || 0, id: match.id, metadata: match.metadata as Record<string, string> | undefined, }; }); // Sort by relevance score (highest first) memories.sort((a, b) => b.score - a.score); return memories; } /** * Updates a memory vector embedding * @param memoryId - ID of the memory to update * @param newContent - New content for the memory * @param userId - User ID to associate with the memory (used as namespace) * @param env - Environment containing Vectorize service */ export async function updateMemoryVector( memoryId: string, newContent: string, userId: string, env: Env, metadata: Record<string, string> = {} ): Promise<void> { // Generate new embedding const newValues = await generateEmbeddings(newContent, env); // Upsert into Vectorize to update await env.VECTORIZE.upsert([ { id: memoryId, values: newValues, namespace: userId, metadata: { content: newContent, ...metadata }, // Update metadata as well }, ]); console.log(`Vector for memory ${memoryId} (namespace ${userId}) updated.`); } /** * Deletes a vector by its ID from the Vectorize index * @param memoryId - ID of the memory to delete * @param userId - User ID to associate with the memory (used as namespace) * @param env - Environment containing Vectorize service */ export async function deleteVectorById( memoryId: string, userId: string, env: Env, ): Promise<void> { try { const result = await (env.VECTORIZE as any).deleteByIds([memoryId], { namespace: userId, }); console.log( `Deleted vector ID ${memoryId} from Vectorize namespace ${userId}. Result:`, result, ); } catch (error) { console.error( `Error deleting vector ID ${memoryId} from Vectorize namespace ${userId}:`, error, ); throw error; } }